A Cytomine software for classifying crops with Random subwindows and ExtraTrees.
All images and related data are downloaded to the /data folder of the container.
The final model object is serialized and compressed using joblib.dump with compression level of 3 and attached to the job.
The compressed model is either a PyxitClassifier (--svm set to false) or a SvmPyxitClassifier (--svm set to true).
Also attached as job properties:
classes: the sorted ids of the terms that were used to train the classifier. Indexes of this table are the actual classes used by the classifier.positive_classes: if the problem was binarized, the ids of the terms used as positive
This software is a contribution from the ULiège Cytomine research team